CirrusBench: Evaluating LLM-based Agents Beyond Correctness in Real-World Cloud Service Environments | ScienceToStartup | ScienceToStartup